(IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 11, No. 12, 2020 Industrial Energy Load Profile Forecasting under Enhanced Time of Use Tariff (ETOU) using Artificial Neural Network

Mohamad Fani Sulaima1, Siti Aishah Abu Hanipah2, Nur Rafiqah Abdul Razif3 Intan Azmira Wan Abdul Razak4, Aida Fazliana Abdul Kadir5, Zul Hasrizal Bohari6 Faculty of Universiti Teknikal Malaysia Melaka Melaka, Malaysia

Abstract—The program involves consumers two parties' needs: generation and consumer sides, to mitigate and reducing global CO2 emission. In respectively. sustaining this effort, energy provider such as Tenaga Nasional Berhad (TNB) in Peninsular Malaysia has introduced Enhance The TOU method design considering both consumers and Time of Use (ETOU) tariff. However, since 2015, small numbers generation was presented in [4], [5]. The founding is reliable, join the ETOU program due to less confidence in managing their where the price elasticity consideration has been given energy consumption profile. Thus, this study provides an attention. The marginal cost of the generation comes to the optimum forecasting load profile model for TOU and ETOU minimum profit rate is secured. Simultaneously, the promotion tariffs using Artificial Neural Network (ANN). An industry's of the consumers' participants to the TOU program was a average energy profile has been used as a case study, while the critical factor that was always being discussed. The TOU tariff forecasting technique has been conducted to find the optimum design is also close to the residential consumers where the energy load profile congruently. The load shifting technique has behavior and the appliances' arrangement to involve in load been adopted under ETOU tariff price while integrating to the management is the command issue. Optimal load arrangement ANN procedure. A significant comparison in terms of cost has been practiced while reflecting the energy provider or reduction between TOU and ETOU electricity tariffs has been retailers' TOU tariff as explained in [6]. The consumers' made. In contrast, ANN performance results in searching for the response is essential for the TOU tariff's sustainable design, best-shifted load profile have been analyzed accordingly. From where the constitution of the demand response program toward the proposed method, the total electricity cost saving has been human factor contribution is crucially needed. founded to be saved for about 7.9% monthly. It is hoped that this work will benefit the energy authority and consumers in future In Peninsular Malaysia, is facing dramatic changes in the action, respectively. Malaysia Energy Supply Industry (MESI), electricity companies need to examine associated business models and a Keywords—Time of use; artificial neural network; energy host of potential strategies to solve equations for over forecasting; load profile electricity demand from consumers especially in next MESI I. INTRODUCTION 2.0 [7]. TNB introduced many schemes, including ETOU tariffs, to benefit the commercial and industrial consumers, but Demand Side Management (DSM) consists of Demand less of them could implement load shifting and join the Response (DR) program to promote a better independent load program correctly. management strategy for the consumers in dealing with the pricing and the time allocation. Under the DR framework, there Thus, in this paper, the analysis and suggestion for the is a price-based program related to time-based price. Under this ETOU tariff study to deal with industrial load profile is structure, the consumers will select a program correlated to explored. Meanwhile the implementation of the artificial their energy load profile management. At the same time, they algorithm has been given attention to produce the optimal can shift the most available load to the lower price rate at a profile of the energy to meet the lower cost of the electricity particular time zone [1]. One of the commands price-based under ETOU program. The related works have been presented programs is the Time of Use (TOU) tariff. in Section II. Meanwhile, the detailed explanation of the specific industrial load profile formulation under the ETOU The TOU tariff has been proposed with various designs and tariff price has been written in Section III. Meanwhile, depending on the national policy implemented through energy Section IV introduced ANN's implementation in processing the providers' tariffs to the consumers. The authors in [2] have load profile forecasting under the ETOU tariff price. The highlighted the TOU time zones by the region. For example, simulation results comparing the actual load profile and the India's command design was four zones, and China was five- optimization load profile are demonstrated in Section V. The zones and Brazil with three zones. However, the reference [3] last Section VI concludes the overall study contribution and presents a different opinion where the arrangement of the time recommendation for an available future research opportunity. zones design in TOU should be flexible enough to comply with

204 | Page www.ijacsa.thesai.org (IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 11, No. 12, 2020

II. RELATED WORKS The average load profile versus time (energy consumption) Unsuitable tariffs for different load profiles will increase for every peak are plotted. The reference formula was obtained energy costs and lead to the DR program's wrong perception from the graph by performing regression, and data analysis [8]. In [9], [10], the ETOU tariff price has been adopted to the processes have been shown in the results. Remodel the commercial and industrial consumers' reference energy profile. reference formula by substituting the total peak cost and rated The finding has been analyzed where the flexibility of the load peak value to find a new peak and off-peak ratio. Thus, the shifting weightage has been constructed accordingly. There total off-peak cost is written as YTOP: were limitations found that the minimum load shifting = × ; ( ) (2) weightage was higher for the consumers to gain the cost reduction. 푌Total푇푂푃 peak푃푇푂푃 cost, 푅Y푇푂푃TP: 푀푌푅 The previous study's suggestion contributes to = × ; ( ) (3) implementing the optimization algorithms to find the optimal 푌푇푃 The푃 푇푃formulas푅푇푃 for 푀푌푅the total cost for every peak are designed load profile that reflects the ETOU tariff price focusing on the by using new ratio values. Verify the total formula cost for Malaysian condition. The application of the peak and off-peak by plotting new graph time versus power Particle Swarm Optimization (PSO) algorithm has been using every peak formula's total cost. Perform data analysis presented in reference [11] to find the appropriate load and regression to get the formula from the graph. Compared categories under six groups of the sectors in peninsular the ratio value with the calculated ratio value from (4). The Malaysia. The powerful strategies to reduce total cost for TOU tariff, T is: total electricity under ETOU tariff have been decreased by T using Ant Colony Optimization (ACO) [12] and [13]. The = × × ; ( ) (4) authors in [2] present an Evolutionary Algorithm (EA) as the optimal search of load profile in tackling the ETOU price. The 푇Where;푇 퐶푇푂푃 퐶푇푃 퐶푇푀퐷 푀푌푅 future recommendation has been made in the field of CTMD = total electricity cost for maximum demand knowledge where the optimization algorithm's impact would be further studied to load profile forecasting under the ETOU CTOP = total electricity cost for off-peak time tariff price. CTP = total electricity cost for peak time The application of a conventional vector machine in Repeat all Eq. (1) until (4) for ETOU tariff, but the determining the load forecasting was critically discussed in arrangement of the mid-peak cost follows Eq. (3). Thus, the [14]. Combining the optimization algorithm produced more total cost for ETOU tariff, T is: impact in reducing the root means square value while updating E the convergence time, as proven by [15]. As different studies = × × × ; ( ) (5) reported of the application of ANN-Self Organizing Mapping for the load forecasting, the significant analysis made has 푇Where;퐸 퐶퐸푂푃 퐶퐸푀푃 퐶퐸푃 퐶푇푀퐷 푀푌푅 contributed to the excellent summary of the group clustering CTMD = total electricity cost for maximum demand for the electricity profiles as the example in [16]–[18]. However, those references less reflect the tariff structure in CEOP = total electricity cost for off-peak time considering the TOU's impact on benefiting consumers of the C = total electricity cost for medium peak time electricity cost reduction concurrently. EMP C = total electricity cost for peak time Regarding the load profile forecasting, which reflects EP ETOU electricity tariff in peninsular Malaysia, there is no The TT and TE formula is tested with a conventional particular study to date for the best of the knowledge. Thus, in method by substituting all the ratio, Q and tariff rates, R values, this study, the mathematical equation stage for ETOU price has and formula value. To observed TOU and ETOU tariff, graph been explored and explained. In contrast, the effective tariff versus cost TOU and ETOU are performed and analyzed. formulation for optimal load profile forecasting and the ANN's implementation has been proposed congruently. The significant Hence, the optimal formulation for daily energy case study of an industry load profile from the medium voltage consumption cost (EC) is Eq. (6). category was chosen with the operation time was 24 hours. In = ( ) ( ) (6) contrast, the optimal load profile results to reduce the 48 electricity cost have been explained accordingly. Where;퐸퐶 ∑ R(t)푡=1 푅 = 푡ETOU퐿 푡 rate of each hour III. FORMULATION OF INDUSTRIAL ENERGY LOAD L(t) = Load of each 30 minutes (0 < 48) FORECASTING UNDER ETOU TARIFF PRICE Since the ETOU tariff price is fix푡ed≤ every day, consumers The flow of the ETOU mathematical stage for demand may shift their loads to reduce energy costs. Then the load pricing is presented as follows: Based on the average load shifting reflect the energy cost, ECR could be written as profile data, the range for peak time, off-peak time, and = ( ) ( ) (7) maximum demand are identified. Maximum demand cost, 48 CTMD: 퐸Where;퐶푅 ∑EC푡=R1 푅= Energy푡 퐿푅 푡 cost of that day under load shifting; = × ; ( ) (1)

퐶푇푀퐷 푃푀퐷 푅푇푀퐷 푀푌푅 205 | Page www.ijacsa.thesai.org (IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 11, No. 12, 2020

LR(t) = Load shifting of each 30 minutes Where; Xn = Normalized data;

(0 < 48); Xj = Actual data;

The푡 ≤ industrial consumers' objective function to enjoy Xmax = Maximum value in actual data; minimum energy cost by load shifting is written in Eq. (8) accordingly. Xmin = Minimum value in actual data; ( ) = min [ ( ) ( )] (8) C. Data Testing 48 After training, the maps are tested with testing data of 푚푖푛The퐸퐶 푅total energy∑푡 =consumption1 푅 푡 퐿푅 푡 in a day for baseload or energy profile and energy cost. This method is to associate the actual load and after load shifting must be equal has been set as energy profile and tariff simultaneously. The data testing the constraint. The consumers may only change some of their process will obtain the pattern for the most suitable load loads from the high price periods to the low-price periods. But shifting profile. the total consumption maybe not changed due to similar behaviors in their daily life or production. However, in the D. Data Forecasting condition of the simulation forecasting process, if the load After testing, all the load data will be forecasted. The curve shifting condition less than 5% from the actual load, the results of energy profile forecasting and actual profile is obtained by can also be considered for comparison. Thus, the constrain the simulation process accordingly. condition could be written as: V. RESULTS AND ANALYSIS ( ) ( ) (9) 48 48 Fig. 1 shows the energy profile, which is having different ∑푡=1IV.퐿푅 I푡MPLEMENTATION≈ ∑푡=1 퐿 푡 OF ARTIFICIAL NEURAL NETWORK power consumes during every time zone. The industrial Table I shows stages that are involved in load forecasting electricity installation is registered under E2 tariff while and optimization by using ANN. Meanwhile, the explanation comparison has been made to E1 flat and E1 ETOU tariff of the implementation has been written in sub-section A until E price. The energy pattern for January 25, 2016 (Thursday- accordingly. Working day) showed that the energy was fully consumed during peak time zones. The Red dot on the figure indicates TABLE I. ANN PROCESS TO DETERMINE THE OPTIMAL LOAD PROFILE that the max power consumes 3,330kW, which will be UNDER ETOU TARIFF PRICE considered maximum demand. The maximum demand position is at 3:00 PM, which is in the peak time zone for TOU and Stages Process ETOU tariffs. Fig. 2 shows the energy consumed profile for an Step 1 Data Organization industrial within 14 days (working day). The profiles demonstrate power consumption for working days is just the Step 2 Data Training same, which is fully consumed during normal working hours, Step 3 Data testing and only 50% operate at night. For the normal 24 hours Step 4 Data Forecasting operation in the manufacturing batch process, the repeating profile would be expected. The load profile index would be A. Data Organization calculated as well to see the level of the significant correlation The previous energy profile data is used as input data. The of the maximum demand and the energy consumption. data include daily energy profile value in minutes in January The best prediction of the minimum energy cost pattern 2016. Twenty-eight data sets are divided into two main groups, was produced in Fig. 3. The blue line refers to the actual which are x_(LoadProfile) and y_EnergyCost. An energy cost with ETOU tariff, while the red line shows the new x_(LoadProfile) is industry energy profile for 14 data set for energy cost after optimization with a regression of 0.7003. The working days. Meanwhile, y_EnergyCost is for total energy graph prediction graph showed various changing patterns to cost per day with ETOU tariff for 14 data set for working days. reach the minimum energy cost. Since the ETOU tariff has B. Data Training been divided by Six-Segmentation of the zone that reflect three prices unit, the investigation could be analyzed based on that In the training process, the data structure must be points. The simulation power profile for the ETOU tariff on the normalized. The normalization is copied to the map structure mid-peak and peak zones have followed the baseline profile. during the trained ANN. The data input and output will However, the consumption of the electricity has reduced normalize and scale the variable values between negative and significantly. The power consumption on off-peak zone from positive ones. The optimum number of neurons must be the 10:00PM to the 8:00AM has increased tremendously to considered during normalization. The data are normalized by show the performance of the algorithm to find the optimal load using the following formula: to be shifted from peak to off-peak zone. Thus, all those = ÷ ( ) (10) condition during the adaptation of the ANN has contributed to 푥푚푎푥+푥푚푖푛 푥푚푎푥−푥푚푖푛 the cost profile such presented in Fig. 3 accordingly. 푥푛 �푥푗 − � 2 �� 2

206 | Page www.ijacsa.thesai.org (IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 11, No. 12, 2020

Fig. 1. Energy Profile in every 30 minutes for 24 hours from 12:30 am until 12:00am (average).

Fig. 2. The Energy Pattern of Industrial Sector for 14 days.

Fig. 3. The Energy Cost Pattern for Actual Data and Prediction Data.

207 | Page www.ijacsa.thesai.org (IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 11, No. 12, 2020

Fig. 4. The Energy Profile Pattern for Actual Data and Optimization Data.

The optimal load response of typical industrial load to be shifted to off-peak hour. Thus, the peak demand movement ETOU power price for demand-side management are shown in condition contributes to effort the efficiency management of Fig. 4. The customer shifts some of the loads from the high the generation side where the critical type of the generation price period to the low-price period in order to achieve would not be run in the long time every day. minimum energy cost in the day. The load is reduced about 10% at the price peak (about 3:00 PM), and 15% reduction at TABLE II. THE COST COMPARISON BETWEEN ACTUAL ENERGY PROFILE mid-peak (about 11:30 AM). Meanwhile, the load is increased AND OPTIMIZATION ENERGY PROFILE by about 20% at the price low peak (about 3:00 AM) and 15% Actual Case Optimization Records Reduction improved at low peak (about 10:00PM). Overall load E1 flat Case E1 ETOU optimization in average ratio was 40:60 of peak and off-peak zone, respectively. Daily Energy cost 40,201.00 39,369.52 2.07%

Maximum demand The maximum demand position also shifted from peak 3,330 (peak) 3,210 (mid peak) 3.60% region to mid peak region while the peak demand was (kW) allocated in off-peak concurrently. It is indicated that the peak Maximum demand cost 98,568.00 95,016.00 3.60% electricity consumption is reduced, and the off-peak electricity (MYR) consumption increases significantly, which affects the power Total Monthly 1,304,598.00 1,201,145.44 7.90% system's regular operation. Table II shows the reduction of the Electricity Cost (MYR) analysis for the energy and maximum demand cost based on two types of tariffs offered to the consumer. Since the ETOU TABLE III. THE COMPARISON OF ENERGY COST BY USING OPTIMIZATION offers off-peak price for the weekend, the advantage has gained ENERGY PROFILE to the industrial profile when running the same operation every Without Optimization With Optimization day. Total forecasting monthly electricity bill when using the Tariff Monthly Energy Cost, Monthly Energy Saving (%) optimization method was reduced up to 7.9% which (MYR) Cost, (MYR) approximately MYR 103,453.00 or yearly saving for about TOU 1,262,019.00 1,194,424.80 5.36 MYR 1,241,436.00. ETOU 1,318,238.00 1,201,145.44 8.80 Besides, Table III presents the comparison of the cost reduction between the case of optimum TOU tariff and case of VI. CONCLUSION the optimum ETOU tariff price accordingly. Both of the price rates has been used in the simulation. Meanwhile, the optimum In this study, the ANN has been applied to forecast the load profile was defined as explained before. Through the optimum load profile under TOU and ETOU tariff prices. The comparison, both TOU and ETOU are produced excellence comparison of the both tariffs has been done where the benefit cost reduction for approximately 5%~8%, but it would be of the TOU and ETOU price could be received by consumers recommended that the ETOU tariff scheme be able to benefit were quantified based on the monthly and yearly results. the consumers with the condition that the optimum load must Meanwhile, the significant is shifted load must converge in the be applied. The study's cost-saving recorded shows the zone of the mid-peak for the lower charge of the maximum significant contribution of the demand response program to the demand. As the investigation statement presents that the ETOU consumers. In contrast, the peak demand would be projected to can be optimum adopted by the consumers where the minimum

208 | Page www.ijacsa.thesai.org (IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 11, No. 12, 2020 load shifting condition is applied. Besides, the electricity bill's [7] Energy Commission (Malaysia), “Reimaging Malaysia Energy Supply cost can be reduced; the peak demand has been shifted to the Industry (MESI 2.0),” 2019. off-peak zone that brings benefits to the generation side. The [8] S. Mohajeryami, I. N. Moghaddam, M. Doostan, B. Vatani, and P. Schwarz, “A novel economic model for price-based demand response,” future recommendation of the research continuity can be Electr. Power Syst. Res., vol. 135, pp. 1–9, 2016. explored on the comparison of the forecasting algorithms by [9] N. A. M. Azman, M. Pauzi Abdullah, M. Y. Hassan, and D. M. Said, using the other techniques while using a multi-objective “Enhanced Time of Use Electricity Pricing for Industrial Customers in environment. Meanwhile other factor of the effect toward load Malaysia,” Indones. J. Electr. Eng. Comput. Sci., vol. 6, no. 1, pp. 155– factor index and the energy economic side should be 160, 2017. considered as well. [10] N. A. M. Azman, M. P. Abdullah, M. Y. Hassan, D. M. Said, and F. Hussin, “Enhanced time of use electricity pricing for commercial ACKNOWLEDGMENT customers in Malaysia,” Pertanika J. Sci. Technol., vol. 25, pp. 285–294, 2017. The authors would like to thank Universiti Teknikal [11] M. F. Sulaima, N. Y. Dahlan, Z. M. Yasin, M. M. Rosli, Z. Omar, and M. Malaysia Melaka (UTeM) for all the support given. The Y. Hassan, “A review of electricity pricing in peninsular Malaysia: publication of the paper is sponsored under the publication Empirical investigation about the appropriateness of Enhanced Time of support scheme under Centre for Research and Innovation Use (ETOU) electricity tariff,” Renew. Sustain. Energy Rev., vol. 110, UTeM (CRIM). pp. 348–367, 2019. [12] M. Sulaima, N. Dahlan, and Z. Yasin, “Effective Electricity Cost REFERENCES Management in a Manufacturing Operation by Using Optimal ETOU [1] F. Meyabadi and M. H. Deihimi, “A review of demand-side management: Tariff Formulation,” Int. J. Electr. Electron. Syst. Res., vol. 15, pp. 82– Reconsidering theoretical framework,” Renew. Sustain. Energy Rev., vol. 93, 2019. 80, no. January 2016, pp. 367–379, 2017. [13] M. F. Sulaima, N. Y. Dahlan, M. H. Isa, M. N. Othman, Z. M. Yasin, and [2] M. F. Sulaima, N. Y. Dahlan, Z. M. Yasin, N. A. M. Asari, and Z. H. H. A. Kasdirin, “ETOU electricity tariff for manufacturing load shifting Bohari, “Optimum enhance time of use (ETOU) for demand side strategy using ACO algorithm,” Bull. Electr. Enginnering Informatics, electricity pricing in regulated market: An implementation using vol. 8, no. 1, pp. 21–29, 2019. evolutionary algorithm,” Indones. J. Electr. Eng. Comput. Sci., vol. 8, no. [14] D. Niu, Y. Wang, and D. Dash, “Power load forecasting using support 1, 2017. vector machine and ant colony optimization,” Expert Syst. Appl., vol. 37, [3] V. Venizelou, N. Philippou, M. Hadjipanayi, G. Makrides, V. Efthymiou, pp. 2531–2539, 2010. and G. E. Georghiou, “Development of a novel time-of-use tariff [15] A. Ghasemi, H. Shayeghi, M. Moradzadeh, and M. Nooshyar, “A novel algorithm for residential prosumer price-based demand side hybrid algorithm for electricity price and load forecasting in smart grids management,” Energy, vol. 142, pp. 633–646, 2018. with demand-side management,” Appl. Energy, vol. 177, pp. 40–59, [4] M. Pauzi Abdullah, M. N. Nazatul Shiema, M. Y. Hassan, and F. Hussin, 2016. “Optimizing Time of Use Electricity Pricing in Regulalted Market,” J. [16] N. . Z. N. F. . G. N.N Atira; I. Azmiara, Z.H. Bohari, “Medium Term Teknol., vol. 7, pp. 31–39, 2013. Load Forecasting Using Statistical Feature Self Organizing Maps ( SOM [5] N. S. M. Nazar, M. P. Abdullah, M. Y. Hassan, and F. Hussin, “Time- ),” J. Telecommun. Electron. Comput. Eng., vol. 11, no. 2, pp. 25–29, based electricity pricing for Demand Response implementation in 2019. monopolized electricity market,” in SCOReD 2012 - 2012 IEEE Student [17] X. Lin, Z. Tian, Y. Lu, H. Zhang, and J. Niu, “Short-term forecast model Conference on Research and Development, 2012, pp. 178–181. of cooling load using load component disaggregation,” Appl. Therm. [6] S. Yilmaz, S. Weber, and M. K. Patel, “Who is sensitive to DSM ? Eng., vol. 157, pp. 1–13, 2019. Understanding the determinants of the shape of electricity load curves [18] Z. H. Bohari, H. S. Azemy, M. N. M. Nasir, M. F. Baharom, and M. F. and demand shifting : Socio-demographic characteristics, appliance use Sulaima, “Reliable Short Term Load Forecasting Using Self Organizing and attitudes,” Energy Policy, vol. 133, pp. 1–13, 2019. Map (SOM) in Deregulated Electricity Market,” J. Theor. Appl. Inf. Technol., vol. 79, no. 3, pp. 389–394, 2015.

209 | Page www.ijacsa.thesai.org